Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add jxoesneon/Ciel --skill double-loopgit clone --depth 1 https://github.com/jxoesneon/CielWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/jxoesneon/ciel/double-loop)<a href="https://agentmods.dev/skills/jxoesneon/ciel/double-loop"><img src="https://agentmods.dev/badge/skills/jxoesneon/ciel/double-loop/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/jxoesneon/ciel/double-loop"><img src="https://agentmods.dev/badge/skills/jxoesneon/ciel/double-loop.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00060 | $0.00987 |
| Opus 5.5 | $0.00024 | $0.00395 |
| Sonnet 5.5 | $0.00012 | $0.00197 |
| Haiku 4.5 | $0.00006 | $0.00099 |
Grade A, and why
double-loop scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 24d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CIEL ADAPTATION: Double Agentic Loop
Two nested loops on different cadences, separated by a single structured channel:
- Outer loop (Supervisor — this session): decomposes the goal into file-disjoint Work Packets, maintains the slot pool, independently verifies every returned result (never accepts the worker's own say-so), issues Correction Packets on failure, and replans when mid-run evidence changes the picture. The outer loop holds all global state: the worklist, the file-ownership map, the integration order.
- Inner loop (Worker — one ephemeral subagent per packet): runs its own tool-use loop on exactly one packet — implement, self-verify, report a Structured Result. The inner loop holds only local state. It never sees the whole plan.
The loops are connected by two channels only: the packet going down, and the structured result coming up. The interrupt channel is the worker's BLOCKED/NEEDS_CONTEXT report plus the supervisor's own file-state reads.
The Core Protocol
Pre-dispatch hooks (run once per activation)
- Worklist freeze: enumerate every open item; assign each a stable WP id.
- File-ownership map: for every WP, list the exact files it may touch. Two WPs may not run concurrently if their file sets intersect. Shared files force sequencing or merging into one WP.
- Tree-state check:
git statusclean or accounted-for; note in-flight work from other sessions and route around it. - Wave plan: order WPs so sweeping refactors (cross-cutting type/DI changes) run LAST, serially, after all localized packets land.
Per-packet hooks
- Packet author hook: every packet carries — scope, exact target files, acceptance criteria, evidence citations (file:line), commit-message convention, and the "NO PUSH / NO unrelated files" rule. Pass only what the worker needs.
- Completion hook (mandatory independent verification): on each worker return —
git log/git diff— the commit exists and touches ONLY the packet's files.- Run the packet's own verification evidence (the tests it claims) — spot-run them yourself; do not trust the report alone.
- Confirm acceptance criteria against the actual diff.
- Close the tracked issue with a resolution comment citing the commit.
- Then — and only then — hot-swap the freed slot to the next packet.
- Failure hook: on
BLOCKED/test failure/partial work — diagnose from the diff, author a Correction Packet (narrow, specific, referencing the failing evidence), re-dispatch the SAME slot. Never let a failed packet silently drop. - Integration hook: after each wave, run the affected test files together and
dart analyze(or project equivalent) across the union of changed files — cross-packet conflicts surface here, not in the worker loops.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 24d ago First seen · 59 lines · 60 tokens per session scan A 71f74b8bd89e
double-loop is a skill published in the GitHub repository jxoesneon/Ciel (1 stars, last pushed today), licensed Apache-2.0. It adds 60 tokens to every session and 987 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-24.
Other skills, from other repositories
factory-handoff
Hand one exact agent-ready ticket to the central Factory runtime over the configured SSH forced-command boundary. Use when an operator asks to hand off an existing ticket or a prose request instead of implementing it in the current session.
ticket-spec
Turn a raw Linear Triage ticket into a fully specified ai:agent-ready one by exploring the codebase. Use when specifying, refining, or promoting tickets, or when asked why a ticket isn't dispatchable.
feature-flow
Cria o diretório canônico da task (docs/todo/NNN-slug/), a branch e o scaffolding inicial. Use para "criar feature", "iniciar tarefa" ou "abrir task".
task-tracker
Escaneia docs/todo/ reportando o Status de cada task e arquiva em docs/done/ somente as entregues. Use para "verificar tasks", "o que está pendente" ou "arquivar concluídas".
autonomous-project-manager
This is the 'Master Blueprint' for long-running autonomous tasks. Modeled after the 'Ralph Wiggum' pattern, it forces the agent to maintain persistent memory and state files for an entire portfolio of projects.
feature-retirement-comms-planner
Identifies users dependent on a feature being removed. Drafts a timeline of 3 emails: 'Heads up', 'Alternative Workflows', and 'Final Reminder'.